memory leak in mbedtls using net_app DTLS client

[mazhenke]

I am using the DTLS client in zephyr, and I found there is memory leak when run the following flow mutli times:

1. net_app_init_udp_client
2. net_app_set_cb
3. net_app_client_tls
4. net_app_connect
5. net_app_close
6. net_app_release
7. go to step1 and repeat

Seems the mebedtls_ssl_xxx_init functions in _net_app_tls_init function doesn't release the heap when calling net_app_release

[mateusz-]

I think I am also running into this. If I call mqtt_connect a couple of times I end up getting an error in my setup cert callback. Specifically, mbedtls_x509_crt_parse_der fails with MBEDTLS_ERR_X509_ALLOC_FAILED.

My callback is straight out of the mqtt example file.

[mike-scott]

Hello,

On 06/28/2018 11:50 PM, Zhenke Ma wrote: I am using the DTLS client in zephyr, and I found there is memory leak when run the following flow mutli times: 1. net_app_init_udp_client 2. net_app_set_cb 3. net_app_client_tls 4. net_app_connect 5. net_app_close 6. net_app_release 7. go to step1 and repeat Seems the mebedtls_ssl_xxx_init functions in _net_app_tls_init function doesn't release the heap when calling net_app_release

I have a WIP patchset which includes this hack to fix the leak you're describing.  I need some time to submit an actual fix: 82c5d82#diff-88ec0070db0157cb49aeb6609e210f1aR3883 Not sure if knowing where the leak is helps, but always good to get more information. NOTE: Added Jukka for reference - Mike

…

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#8631>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AI7BB8loGMuLiJ6JKwyGTaN4C2b5iHMZks5uBc5IgaJpZM4U8hm8>.

[pfalcon]

Similar comment to what was posted to #8746 :

Recently, DTLS socket support was merged: #9338 , and that's going to be the new official API for (D)TLS usage. You are thus recommended to switch to it whenever possible. You may find that older net_app DTLS client is in "maintenance mode" (that actually depends on developers' time availability, and AFAIK, they're fully occupied). I thus lower priority of this ticket.

[pfalcon]

@mike-scott :

I have a WIP patchset which includes this hack to fix the leak you're describing. I need some time to submit an actual fix: 82c5d82#diff-88ec0070db0157cb49aeb6609e210f1aR3883

Definitely please submit a fix if you have that investigated. Thanks.

[tautologyclub]

I'm guessing all net_app-dtls issues will be more or less left to hang henceforth?

That's understandable I guess, but I'd make sure to mark it as __deprecated if so.

We discovered a really weird bug today where net_app registered a rx callback which was called by one of the handshake functions, but the rx callback itself indirectly called the handshake function itself - thus recursing itself (silently, for some reason) into oblivion.

Our current intuition says that we should probably forget about net_app and migrate to sockets because it seems like net_app is a lot more __deprecated than the documentation suggests. True?

[jukkar]

net-app is gone, so closing this issue